New research tackles LLM and VLM hallucinations with advanced detection methods
ByPulseAugur Editorial·[13 sources]·
Researchers are developing new methods to combat hallucinations in large language models (LLMs) and vision-language models (VLMs). One approach, "Verify when Uncertain," uses cross-model consistency checking to improve hallucination detection accuracy while reducing computational costs. Another method, CORTEX, focuses on token-level detection in retrieval-augmented generation (RAG) by comparing internal model representations with and without retrieved documents. For VLMs, techniques like ViPSy and Context-aware Attention Intervention (CAI) aim to enhance visual faithfulness by better aligning model outputs with image content and reducing reliance on linguistic priors.
AI
IMPACT
Advances in hallucination detection and mitigation are crucial for increasing the reliability and trustworthiness of LLMs and VLMs in sensitive applications.
RANK_REASON
Multiple academic papers published on arXiv detailing novel methods for detecting and mitigating hallucinations in LLMs, VLMs, and RAG systems.
arXiv:2607.00895v1 Announce Type: new Abstract: Hallucination detection for retrieval-augmented generation (RAG) is usually evaluated on natural-language document evidence. However, grounded generation systems increasingly rely on structured inputs: source code, developer-tool ou…
arXiv:2607.00158v1 Announce Type: new Abstract: Hallucination remains one of the central obstacles to deploying medical LLMs. Yet, even when hallucination can be detected, it is still unclear whether the internal representations associated with it can be used for control rather t…
Hallucination detection for retrieval-augmented generation (RAG) is usually evaluated on natural-language document evidence. However, grounded generation systems increasingly rely on structured inputs: source code, developer-tool output, markdown documents, tables, and repository…
arXiv:2606.31033v1 Announce Type: new Abstract: In this paper, we propose CORTEX, a token-level hallucination detection method for Retrieval-Augmented Generation (RAG). In long-form RAG outputs, hallucinations often arise in localized spans rather than throughout an entire respon…
arXiv cs.AI
TIER_1English(EN)·Yihao Xue, Kristjan Greenewald, Youssef Mroueh, Baharan Mirzasoleiman·
arXiv:2502.15845v2 Announce Type: replace-cross Abstract: Large Language Models (LLMs) often hallucinate, limiting their reliability in sensitive applications. In black-box settings, several self-consistency-based techniques have been proposed for hallucination detection. We empi…
Hallucination remains one of the central obstacles to deploying medical LLMs. Yet, even when hallucination can be detected, it is still unclear whether the internal representations associated with it can be used for control rather than detection alone. Using four open-source mode…
arXiv:2606.28520v1 Announce Type: cross Abstract: Large vision-language models (LVLMs) are increasingly used for clinical image understanding, yet they remain vulnerable to \emph{hallucinations}--producing textual findings or attributes not supported by the image. We present a vi…
arXiv:2606.28401v1 Announce Type: cross Abstract: Vision-Language Models (VLMs) have shown strong performance in visual understanding, yet they still suffer from hallucinations, generating content that is not grounded in the image. Preference alignment is a promising approach to …
In this paper, we propose CORTEX, a token-level hallucination detection method for Retrieval-Augmented Generation (RAG). In long-form RAG outputs, hallucinations often arise in localized spans rather than throughout an entire response. CORTEX therefore identifies ungrounded conte…
We introduce VidPair-Halluc, a new benchmark for evaluating video hallucination in large video models (LVMs) under rigorous and controlled conditions. Unlike previous benchmarks that primarily rely on text-based perturbations or adversarial questions while neglecting the consiste…
arXiv cs.CV
TIER_1English(EN)·Yuqing Lei, Wenbo Lyu, Yingjun Du, Xiantong Zhen, Cees G. M. Snoek, Ling Shao·
arXiv:2606.29847v1 Announce Type: new Abstract: Large Vision-Language Models (LVLMs) excel at multimodal tasks but remain prone to object hallucinations. Prior training-free remedies often uniformly strengthen visual signals, which may also amplify irrelevant regions and introduc…
arXiv:2606.29805v1 Announce Type: new Abstract: Multimodal Large Language Models (MLLMs) are prone to hallucination as their generation preferences are insufficiently calibrated to visual evidence, causing them to fall back on linguistic priors, rather than faithful grounding. In…
<p>The failure mode that scares me most in RAG isn't a wrong answer. It's a confident wrong answer with three citations that don't actually say what the answer claims.</p> <p>So in SWIRL 5 I stopped trusting the model to police itself and added a check that runs <em>after</em> ge…
